Table of Contents

• 1  Research purpose
• 2  Research description
• 3  Data preprocessing
  • 3.1  Columns names correction
  • 3.2  Data and time columns creation
  • 3.3  Interim summary
• 4  Data analysis
  • 4.1  Measuring main parameters (unique users per action etc)
  • 4.2  Division by sessions
  • 4.3  Anomaly detection and elimination
  • 4.4  Interim summary
• 5  Main purpose of the study
  • 5.1  Histogram of events
  • 5.2  Division by events chains
  • 5.3  Impact of events leading to target event by all funnel scenarios
  • 5.4  Timediffer between the prevalent events
  • 5.5  Interim summary
• 6  Hypotheses statement
  • 6.1  First hypothesis check
  • 6.2  Second hypothesis check
  • 6.3  Third hypothesis check
  • 6.4  Interim summary
• 7  Project summary
• 8  Recommendations

Mobile application users behavior analysis

Atachments

• Presentation https://docs.google.com/presentation/d/1PGqQJab22Jl8T3LzK3dtvW4qYUpPVVBiHSoXqHGxvpw/edit?usp=sharing

• Dashboard https://public.tableau.com/views/Project11_DB_2022_11_24/Dashboard1?:language=en-US&publish=yes&:display_count=n&:origin=viz_share_link

Research purpose

• Engagement management enhancement
• Application growth points detection
• User experience improvement
• Profit increase

Research description

Analyze the impact of events on the completion of the target event — viewing contacts

The dataset contains data on events committed in the "Unnecessary Things" mobile application. In it, users sell their unwanted items by posting them on a bulletin board.

---

The dataset contains users data that was performed in the application for the first time after October 7, 2019.

Columns in mobile_sources.csv:

• "userId" — user ID,
• "source' — the source from which the user installed the application.

Columns in mobile_dataset.csv:

• "event.time' — the time of the transaction,
• "user.id " — user ID,
• "event.name " — user action.

Types of actions:

• "advert_open' — opened the ad cards,
• "photos_show' — viewed photos in the ad,
• "tips_show" — saw recommended ads,
• "tips_click" — clicked on the recommended ad,
• "contacts_show" and "show_contacts' — looked at the phone number,
• "contacts_call' — called the number from the ad,
• "map" — opened the ad map,
• "search_1"—"search_7" — various actions related to site search,
• "favorites_add" — added the ad to favorites.

Data preprocessing

#
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import plotly.express as px
from datetime import datetime, timedelta
from plotly import graph_objects as go
import os
from plotly.subplots import make_subplots
from scipy import stats as st
import math as mth

pth1 = 'mobile_sources.csv'
pth2 = 'datasets/mobile_sources.csv'

if os.path.exists(pth1):
    sources = pd.read_csv(pth1, sep=',')
elif os.path.exists(pth2):
    sources = pd.read_csv(pth2, sep=',')
else:
    print('Path not found')

pth1 = 'mobile_dataset.csv'
pth2 = 'datasets/mobile_dataset.csv'

if os.path.exists(pth1):
    data = pd.read_csv(pth1, sep=',', parse_dates=['event.time'])
elif os.path.exists(pth2):
    data = pd.read_csv(pth2, sep=',', parse_dates=['event.time'])
else:
    print('Path not found')

#
def info(data):
    print('------------- First 5 lines ------------')
    display(data.sample(5))
    print('------------- Data types ------------')
    display(data.info())
    print('------------- Gaps ------------')
    for element in data.columns:
        if data[element].isna().any().mean() > 0:
            print(element, '-', data[element].isna().sum())
        else:
            print(element, '- None')
    print('------------- Duplicates ------------')
    if data.duplicated().sum() > 0:
        print(data.duplicated().sum())
    else:
        print('No Duplicates');

info(sources)

------------- First 5 lines ------------

	userId	source
1254	c194f363-da79-4019-88ce-4a97466a08bc	yandex
1800	09e3e8fa-8a11-41c5-be3a-ffd275fc6f38	yandex
3198	79910090-aaad-418e-88e0-b9ece782c13b	yandex
3165	c9c58582-ea41-4b0a-925a-a9f9b8eabf30	yandex
261	eb89a159-6d63-4959-8059-1fc44c73c6e6	google

------------- Data types ------------
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4293 entries, 0 to 4292
Data columns (total 2 columns):
 #   Column  Non-Null Count  Dtype 
---  ------  --------------  ----- 
 0   userId  4293 non-null   object
 1   source  4293 non-null   object
dtypes: object(2)
memory usage: 67.2+ KB

None

------------- Gaps ------------
userId - None
source - None
------------- Duplicates ------------
No Duplicates

info(data)

------------- First 5 lines ------------

	event.time	event.name	user.id
61231	2019-10-30 02:33:13.454263	search_2	fdf57c9e-863a-4057-8b4a-782494f7c107
22471	2019-10-16 12:32:07.708267	search_1	f261b81c-57d2-4a9c-a4f5-3153e5f9b503
32298	2019-10-20 09:27:29.155047	tips_show	779317c3-e226-43bd-887f-2307d441b955
61518	2019-10-30 09:56:42.980757	map	b6b8815c-fa73-46af-9df2-a733a8515e88
58353	2019-10-29 09:39:21.827112	tips_show	558b56e9-fc34-438d-9d12-74cd3d864dfa

------------- Data types ------------
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 74197 entries, 0 to 74196
Data columns (total 3 columns):
 #   Column      Non-Null Count  Dtype         
---  ------      --------------  -----         
 0   event.time  74197 non-null  datetime64[ns]
 1   event.name  74197 non-null  object        
 2   user.id     74197 non-null  object        
dtypes: datetime64[ns](1), object(2)
memory usage: 1.7+ MB

None

------------- Gaps ------------
event.time - None
event.name - None
user.id - None
------------- Duplicates ------------
No Duplicates

Columns names correction

data = data.rename(columns={'event.time':'event_time', 'event.name':'event_name', 'user.id':'user_id'})
sources = sources.rename(columns={'userId':'user_id'})
display(data.sample(), sources.sample())
data.to_csv('data_clear.csv')

	event_time	event_name	user_id
57695	2019-10-28 22:20:56.181555	tips_show	6d0e0712-85b4-43de-b8d2-4048540a4e9d

	user_id	source
128	c88b0187-dd82-4a29-a471-b1df72a9aacb	other

Data and time columns creation

data['date'] = pd.DatetimeIndex(data['event_time']).date
data['time'] = pd.DatetimeIndex(data['event_time']).time
data.sample()

	event_time	event_name	user_id	date	time
9174	2019-10-10 21:31:41.301092	tips_show	aa3f2890-770b-443f-8a31-40750ed39bb5	2019-10-10	21:31:41.301092

display(f'First date {data["event_time"].min()}')
display(f'Last date {data["event_time"].max()}')

'First date 2019-10-07 00:00:00.431357'

'Last date 2019-11-03 23:58:12.532487'

Interim summary

There are no Duplicates nor gaps in both datasets.
Column names has been brought to unified style.
Observation period is nearly a month: 2019-10-07 - 2019-11-03

Data analysis

Measuring main parameters (unique users per action etc)

display(f'{data["user_id"].nunique()} Unique users ')

'4293 Unique users '

display(f'{data["event_name"].nunique()} Unique events ')

'16 Unique events '

display('Unique users per event', data.groupby("event_name")["user_id"].nunique().sort_values(ascending=False))

'Unique users per event'

event_name
tips_show        2801
map              1456
photos_show      1095
contacts_show     979
search_1          787
advert_open       751
search_5          663
search_4          474
favorites_add     351
search_6          330
tips_click        322
search_2          242
contacts_call     213
search_3          208
search_7          157
show_contacts       7
Name: user_id, dtype: int64

Let's interchange 7 events "show_contacts" to "contacts_show" as they seem literally single-valued.

data['event_name'] = data['event_name'].replace({'show_contacts':'contacts_show'})
display('Unique users per event', data.groupby("event_name")["user_id"].nunique().sort_values(ascending=False))

'Unique users per event'

event_name
tips_show        2801
map              1456
photos_show      1095
contacts_show     981
search_1          787
advert_open       751
search_5          663
search_4          474
favorites_add     351
search_6          330
tips_click        322
search_2          242
contacts_call     213
search_3          208
search_7          157
Name: user_id, dtype: int64

Division by sessions

The logic behind division by session has been taken as follows:
We assume that session is a flow of events taken by unique user with timediffer between events of one user more then 20 minutes.

sessions = (data.sort_values(['user_id','event_time']).groupby('user_id')['event_time'].diff() > pd.Timedelta('20Min')).cumsum()
data['session_id'] = data.groupby(['user_id', sessions], sort = False).ngroup() + 1
display(data.head(5))

	event_time	event_name	user_id	date	time	session_id
0	2019-10-07 00:00:00.431357	advert_open	020292ab-89bc-4156-9acf-68bc2783f894	2019-10-07	00:00:00.431357	1
1	2019-10-07 00:00:01.236320	tips_show	020292ab-89bc-4156-9acf-68bc2783f894	2019-10-07	00:00:01.236320	1
2	2019-10-07 00:00:02.245341	tips_show	cf7eda61-9349-469f-ac27-e5b6f5ec475c	2019-10-07	00:00:02.245341	2
3	2019-10-07 00:00:07.039334	tips_show	020292ab-89bc-4156-9acf-68bc2783f894	2019-10-07	00:00:07.039334	1
4	2019-10-07 00:00:56.319813	advert_open	cf7eda61-9349-469f-ac27-e5b6f5ec475c	2019-10-07	00:00:56.319813	2

display(f'Out of all {len(data)} events, There are {data.session_id.nunique()} sessions') 

'Out of all 74197 events, There are 10975 sessions'

Anomaly detection and elimination

display(f'Out of {data.session_id.nunique()} sessions, There are {(data.groupby("session_id")["event_name"].count() == 1).sum()} sessions consisting of 1 event') 

'Out of 10975 sessions, There are 2328 sessions consisting of 1 event'

As our goal is to analyze the impact of preceding events on the target event "contacts_show",
we can eliminate sessions consisting of 2 unique events by creating following filters by sessions:

session1 = data.groupby('session_id')['event_name'].nunique().reset_index()
session1 = session1.query('event_name > 2')
len(session1) 

2296

Creation of a new dataset "data2" containing only sessions longer than 1 event and with variety of 2 or more events

data2 = data.query('session_id in @session1.session_id')
data2.head(5)

	event_time	event_name	user_id	date	time	session_id
0	2019-10-07 00:00:00.431357	advert_open	020292ab-89bc-4156-9acf-68bc2783f894	2019-10-07	00:00:00.431357	1
1	2019-10-07 00:00:01.236320	tips_show	020292ab-89bc-4156-9acf-68bc2783f894	2019-10-07	00:00:01.236320	1
2	2019-10-07 00:00:02.245341	tips_show	cf7eda61-9349-469f-ac27-e5b6f5ec475c	2019-10-07	00:00:02.245341	2
3	2019-10-07 00:00:07.039334	tips_show	020292ab-89bc-4156-9acf-68bc2783f894	2019-10-07	00:00:07.039334	1
4	2019-10-07 00:00:56.319813	advert_open	cf7eda61-9349-469f-ac27-e5b6f5ec475c	2019-10-07	00:00:56.319813	2

display(f'Data has been reduced to {len(data2)} from {len(data)}')
display(f'Number of sessions has been reduced to {data2["session_id"].nunique()} from {data["session_id"].nunique()}')

'Data has been reduced to 29492 from 74197'

'Number of sessions has been reduced to 2296 from 10975'

Interim summary

10975 sessions has been allocated.
New dataset 'data2' has been created: containing only sessions longer than 2 unique events and containing target event. In result:
Data has been reduced to 29492 from 74197 and
Number of sessions has been reduced to 2296 from 10975

Main purpose of the study

Histogram of events

#
trace1 = go.Histogram(
    x=data['event_name'],
    opacity=0.75,
    name='Initial Data',
    marker_color='#3A8DF6')
trace2 = go.Histogram(
    x=data2['event_name'],
    opacity=0.75,
    name='Cleared Data',
    marker_color='#92f63a')

graph = [trace1, trace2]
layout = go.Layout(
    title='Frequency of events in initial and cleared datasets',
    barmode='overlay',
    height=900,
    xaxis=dict(
        title='Event name', 
        categoryorder='total descending'),
    yaxis=dict(
        title='Quantity of events',
        overlaying='y'),)

fig = go.Figure(data=graph, layout=layout)
fig.show()

After clearing the data it appeared that the most popular events "tips_show", "photos_show" and "search_1" in 70% of the time is involved into pointless sessions conisting only of the one event.
Directed conscious actions demanding an active action from user like "advert_open", "contacts_show" and "map" has shown less reduction.

Division by events chains

session2 = data2.sort_values(by='event_name').groupby('session_id', as_index=False).agg(event_flow=('event_name', 'unique'))
session2['event_flow'] =  [', '.join(map(str, i)) for i in session2['event_flow']]

funnel = session2['event_flow'].value_counts()
display('Top-10 event steps:', funnel.head(10))

'Top-10 event steps:'

advert_open, map, tips_show                  300
contacts_show, map, tips_show                176
advert_open, map, search_3, tips_show         86
search_4, search_5, search_6, tips_show       80
contacts_show, photos_show, search_1          75
search_5, search_6, tips_show                 75
favorites_add, photos_show, search_1          64
search_4, search_5, tips_show                 63
contacts_call, contacts_show, photos_show     62
contacts_call, contacts_show, search_1        59
Name: event_flow, dtype: int64

#
fig = make_subplots(rows=1, cols=4)

fig.update_layout(showlegend=False,height=800, width=1000, 
                    title=(f'Funnels of Top-4 successful event scenarios in cleared data'), title_font_size = 20 )

session_suc = session2.query('event_flow.str.contains("contacts_show")', engine='python')
funnel_suc = session_suc['event_flow'].value_counts()

for i in range(0,4):
    f = funnel_suc.index[i].split(', ')
    fun = data2.query('event_name == @f').groupby('event_name', as_index=False)['user_id'].nunique().sort_values(by='user_id', ascending=False)
    fig.add_trace(go.Funnel(
    x = fun['user_id'] ,
    y = fun['event_name'],
    
    textposition = "inside",
    texttemplate="%{label}: %{value:} <br> %{percentPrevious}",
    textangle=-90,
    marker = {"color": ["#EBF63A", "#92f63a", "#3A8DF6"]}), 
    row=1, col=i+1 )
    
fig.update_yaxes(visible=False)
fig.show()

Impact of events leading to target event by all funnel scenarios

From Funnel we can see that the most common events leading to "Contacts show" are the following:

• tips_show
• map
• photos_show
• search_1

fig = px.bar( y=funnel[:10].index,  x = funnel[:10].values, color=funnel[:10].index.str.contains('contacts_show'), 
                   color_discrete_sequence=['#3A8DF6', '#92f63a'])
fig.update_layout(title='Frequency of events in cleared data', 
                  yaxis={'categoryorder':'total ascending'},
                  yaxis_title='Quantity of events', 
                  xaxis_title='Event name',
                  legend_title="Hit Target event")
fig.show()

events_main = ['tips_show', 'map', 'photos_show', 'search_1', 'favorites_add', 'advert_open',  'tips_click']
for i in range(len(events_main)):
    x = events_main[i]
    target = "contacts_show"
    a = data2.query("event_name == @x")
    b = data2.query("session_id in @a.session_id and event_name == @target")
    au = a.user_id.nunique()
    bu = b.user_id.nunique()
    display(f'Percent of "{events_main[i]}" events in initial data that lead to target "Contacts show": ')
    display(f'{ bu/au :.3%}')

'Percent of "tips_show" events in initial data that lead to target "Contacts show": '

'25.063%'

'Percent of "map" events in initial data that lead to target "Contacts show": '

'26.039%'

'Percent of "photos_show" events in initial data that lead to target "Contacts show": '

'66.537%'

'Percent of "search_1" events in initial data that lead to target "Contacts show": '

'66.960%'

'Percent of "favorites_add" events in initial data that lead to target "Contacts show": '

'42.661%'

'Percent of "advert_open" events in initial data that lead to target "Contacts show": '

'16.491%'

'Percent of "tips_click" events in initial data that lead to target "Contacts show": '

'37.349%'

---

We can see that the user action "tips_show" and "map" are the most common ways to reach target event.
But this actions make only around 25% conversion to "Contacts show".

It seems that the raise of conversion of "tips_show" and "map" will increase number of "Contacts show" events.

"Photos_show" and "search_1" make around 67% conversion to "Contacts show".

Raise of commiting "photos_show" and "search_1" is also an option, as they show the biggest conversion.

"Favorites_add" makes 42% conversion to "Contacts show".

Raise of commiting "Favorites_add"

Also "tips_click" is showing pretty good conversion -38%, but takes a small part of events flows.

Raise of commiting "tips_click"

"Advert_open" makes the smallest conversion to target event - 15%

Raise of conversion "Advert_open"

Timediffer between the prevalent events

#
session = []
for i in range(len(events_main)):
    x = events_main[i]
    
    session_x = data2.query('session_id in @session_suc.session_id and event_name == @x').groupby('session_id', as_index=False).agg({'event_name':'first','event_time':'min'})
    session_t = data2.query('session_id in @session_x.session_id and event_name == @target').groupby('session_id', as_index=False)['event_time'].min()
    session_x = session_x.merge(session_t, on='session_id')
    session_x['diff'] = session_x['event_time_x'] - session_x['event_time_y']
    session_x['diff'] = session_x['diff'].dt.total_seconds().abs()
    
    display(f'"{events_main[i]}" and "Contacts show" median timediffer:')
    display(f'{ session_x["diff"].median() :.5}, seconds')
    session.append(session_x)
session = pd.concat(session)
display(f'Mean timediffer between prevalent events and target "Contacts show":{ session["diff"].median() :.5}, seconds') 
plt.figure(figsize=(16, 8))
plt.ylim([-1,3000])
print(sns.boxplot(x='event_name', y='diff',  data=session))

'"tips_show" and "Contacts show" median timediffer:'

'278.46, seconds'

'"map" and "Contacts show" median timediffer:'

'342.5, seconds'

'"photos_show" and "Contacts show" median timediffer:'

'306.74, seconds'

'"search_1" and "Contacts show" median timediffer:'

'262.93, seconds'

'"favorites_add" and "Contacts show" median timediffer:'

'277.9, seconds'

'"advert_open" and "Contacts show" median timediffer:'

'260.23, seconds'

'"tips_click" and "Contacts show" median timediffer:'

'404.52, seconds'

'Mean timediffer between prevalent events and target "Contacts show":302.63, seconds'

AxesSubplot(0.125,0.125;0.775x0.755)

Interim summary

We've substracted 5 most popular sessions scenarios:
advert_open, map, tips_show 300
contacts_show, map, tips_show 176
advert_open, map, search_3, tips_show 86
search_4, search_5, search_6, tips_show 80
contacts_show, photos_show, search_1 75

We can see that the user action "tips_show" and "map" are the most common ways to reach target event.
But this actions make only around 25% conversion to "Contacts show".

It seems that the raise of conversion of "tips_show" and "map" will increase number of "Contacts show" events.

"Photos_show" and "search_1" make around 67% conversion to "Contacts show".

Raise of commiting "photos_show" and "search_1" is also an option, as they show the biggest conversion.

"Favorites_add" makes 42% conversion to "Contacts show".

Raise of commiting "Favorites_add"

Also "tips_click" is showing pretty good conversion -38%, but takes a small part of events flows.

Raise of commiting "tips_click"

"Advert_open" makes the smallest conversion to target event - 15%

Raise of conversion "Advert_open"

Mean timediffer between prevalent events and target "Contacts show" is around 300 seconds (5 minutes)

Hypotheses statement

First hypothesis check

H0: Conversion to contact views is equal between two groups: ones who perform tips_show and tips_click actions, and ones with only tips_show.
H1: Conversion to contact views between two groups differs

def z_test(successes1, successes2, trials1, trials2, alpha=0.05):
    p1 = successes1 / trials1
    p2 = successes2 / trials2
    
    p_combined = (successes1 + successes2) / (trials1 + trials2)
    difference = p1 - p2
    
    z_value = difference / mth.sqrt(p_combined * (1 - p_combined) * (1/trials1 + 1/trials2)) 
    distr = st.norm(0, 1)  
    p_value = (1 - distr.cdf(abs(z_value))) * 2 
    print('p-value: ', "%.20f" % p_value )
 
    if (p_value < alpha): 
        display('Reject the null hypothesis, there are statistically significant differences between the samples')
    else:
        display('It was not possible to reject the null hypothesis, there are no statistically significant differences in the samples')

session5 = data.sort_values(by='event_name').groupby('session_id', as_index=False).agg(event_flow=('event_name', 'unique'))
session5['event_flow'] =  [', '.join(map(str, i)) for i in session5['event_flow']]

target1 = 'tips_show'
target2 = 'tips_click'
success = 'contacts_show'

x1 = session5.query('event_flow.str.contains(@target1)', engine='python')
x2 = session5.query('event_flow.str.contains(@target1) and event_flow.str.contains(@target2)', engine='python')

trials1 = data.query('session_id in @x1.session_id')['user_id'].nunique()
trials2 = data.query('session_id in @x2.session_id')['user_id'].nunique()

y1 = x1.query('event_flow.str.contains(@success)', engine='python')
y2 = x2.query('event_flow.str.contains(@success)', engine='python')

successes1 = data.query('session_id in @y1.session_id')['user_id'].nunique()
successes2 = data.query('session_id in @y2.session_id')['user_id'].nunique()

z_test(successes1, successes2, trials1, trials2, alpha=0.05)

p-value:  0.15008551289817906316

'It was not possible to reject the null hypothesis, there are no statistically significant differences in the samples'

Second hypothesis check

H0: Conversion to contact views is equal between group of users who performed favorites_add and who did not.
H1: Conversion to contact views between two groups differs

target = 'favorites_add'
success = 'contacts_show'

x1 = session5.query('event_flow.str.contains(@target)', engine='python')
x2 = session5.query('~event_flow.str.contains(@target)', engine='python')

trials1 = data.query('session_id in @x1.session_id')['user_id'].nunique()
trials2 = data.query('session_id in @x2.session_id')['user_id'].nunique()

y1 = x1.query('event_flow.str.contains(@success)', engine='python')
y2 = x2.query('event_flow.str.contains(@success)', engine='python')

successes1 = data.query('session_id in @y1.session_id')['user_id'].nunique()
successes2 = data.query('session_id in @y2.session_id')['user_id'].nunique()

z_test(successes1, successes2, trials1, trials2, alpha=0.05)

p-value:  0.01336115333748821854

'Reject the null hypothesis, there are statistically significant differences between the samples'

Third hypothesis check

H0: Conversion to contact views is equal between group of users who used photos_show and who did not.
H1: Conversion to contact views between two groups differs

target = 'photos_show'
success = 'contacts_show'

x1 = session5.query('event_flow.str.contains(@target)', engine='python')
x2 = session5.query('~event_flow.str.contains(@target)', engine='python')

trials1 = data.query('session_id in @x1.session_id')['user_id'].nunique()
trials2 = data.query('session_id in @x2.session_id')['user_id'].nunique()

y1 = x1.query('event_flow.str.contains(@success)', engine='python')
y2 = x2.query('event_flow.str.contains(@success)', engine='python')

successes1 = data.query('session_id in @y1.session_id')['user_id'].nunique()
successes2 = data.query('session_id in @y2.session_id')['user_id'].nunique()

z_test(successes1, successes2, trials1, trials2, alpha=0.05)

p-value:  0.69348793142563436298

'It was not possible to reject the null hypothesis, there are no statistically significant differences in the samples'

Interim summary

There are no statistically significant differences in conversion to "contacts_show" between the ones who perform
tips_show and tips_click actions, and ones with only tips_show .

There are statistically significant differences in conversion to "contacts_show" between the users who performed
favorites_add and who did not.

There are no statistically significant differences in conversion to "contacts_show" between the users who performed
photos_show and who did not.

Project summary

10975 sessions has been allocated and has been reduced to 2296 sessions after eliminating double unique event sessions.

---

The most common scenarios of successful event flow are:

• contacts_show, map, tips_show 176
• contacts_show, photos_show, search_1 75
• contacts_call, contacts_show, photos_show 62
• contacts_call, contacts_show, search_1 59

---

"Tips_show" and "map" are the most common ways to reach target event.
But this actions make only around 25% conversion to "Contacts show".

"Photos_show" and "search_1" make around 67% conversion to "Contacts show".

"Favorites_add" makes 42% conversion to "Contacts show".

Also "tips_click" is showing pretty good conversion -38%, but takes a small part of events flows.

"Advert_open" makes the smallest conversion to target event - 15%

---

Mean timediffer between prevalent events and target "Contacts show" is around 5 minutes

---

There are statistically significant differences in conversion to "contacts_show" between the users who performed
favorites_add and who did not.

There are no statistically significant differences in conversion to "contacts_show" between the ones who perform
tips_show and tips_click actions, and ones with only tips_show .

There are no statistically significant differences in conversion to "contacts_show" between the users who performed
photos_show and who did not.

Recommendations

In the order of impact:

1. Raise the conversion of "tips_show" and "map"
2. Increase commission "photos_show" and "search_1"
3. Increase commission "Favorites_add"
4. Increase commission "tips_click"
5. Raise the conversion "Advert_open"